Scaling, White Point & Clipping In An HDR World
A couple weeks ago I published an Insight showing the essentials of HDR setup/workflow with Adobe Premiere Pro and Adobe’s initial support of HDR in their popular NLE.
One of the features I showed was the ‘HDR’ mode of the Lumetri scopes. This feature scales the signal to show bright portions of a shot (100nits +) that might be useful to adjust while grading HDR material. While in HDR mode you get a rather useful nit scale to show you how signal (code values) map to actual light output of a monitor.
This functionality, besides being amazingly useful, shows one of the main tenants of Dolby Vision and ST-2084 HDR – there is a direct correspondence of code value to light output.
I really like this scope, and as I said in that Insight, I wish more scope manufacturers and software designers would considering adding something like it in the future.
After publishing the Insight, I got several emails from members (remember you can always use the comments!) expressing confusion about how to read scopes (particularly the Luma Waveform & RGB Parade) when setup to grade HDR in other applications like Resolve.
You might be thinking – ‘what does that mean? Reading a Waveform or RGB parade is the same no matter where you use one.’
True, but in a tool like Resolve when using RCM to aid you in working with HDR footage, the Waveform & RGB parade can get confusing – particularly when it comes to white level and where clipping may occur.
Same Scope, Just Scaled
If I asked you when grading SDR footage where is black and where is white on the Waveform & RGB Parade in Resolve you’d tell me ‘0 and 1023’ (let’s avoid all the full range/legal range, internal processing etc. talk for this discussion).
That knowledge is burned into your brain, as is how to detect clipping and other essentials of using these scopes.
When you switch Resolve to RCM (Resolve Color Management) and choose any of the ST.2084 (HLG works in a similar fashion for this discussion), Things get weird!
With a 2084 output color space applied, the signal as seen on the Waveform or RGB Parade scales in a pretty dramatic fashion (see above screen shot).
This scaling is normal.
Remember 2084 is based on the PQ curve that Dolby researched and SMPTE endorsed. Without getting into the detailed math, the PQ curve makes a direct correlation between code value and nits.
This is why, as I’ll show you in the video below, that choosing a different NIT output option from the 2084 choices for output color space in Resolve scales the signal accordingly.
One of the most striking ways that this scaling effects monitoring the scopes is with white point clipping.
White Point Clipping
One of the strangest byproducts of this scaling – your white clip point isn’t 1023 anymore! One day it might be (using current scopes) and when there are 10k+ Nit displays. Remember, the specification calls for 10,000 nits for future displays.
When that happens, 1023 in Resolve in a 2084/PQ workflow would map to near 10k Nits (again, yes, I realize things are slightly more complicated with legal/full range mapping and 2084 using legal mapping, but since Resolve is always internally full range and displays that full range on the scopes, it’s a moot point).
For now, in Resolve and using RCM, when you choose say 1000 Nits, your white clip point (because of the scaling described above) changes to 768, and not 1023.
Weird at first, but what I’ve done is simply made a sticky note on my desk to show me the approximate code values for each 2084 Nit output color space setting. Of course, these values could be determined exactly with the PQ math that SMPTE documents in their white paper, but for the purposes of this Insight the approximate values will work just fine:
- HDR ST.2084 300 Nits = White Point Clipping at 640 (approx)
- HDR ST.2084 500 Nits = White Point Clipping at 695 (approx)
- HDR ST.2084 800 Nits = White Point Clipping at 745 (approx)
- HDR ST.2084 1000 Nits = White Point Clipping at 768 (approx)
- HDR ST.2084 2000 Nits = White Point Clipping at 845 (approx)
- HDR ST.2084 4000 Nits (Max Resolve Currently Supports/Dobly Pulsar) = White Point Clipping at 920 (approx)
While these numbers are approximate, they’re easy to see for yourself using the reference level option as I’ll show you in the movie.
You may have also noticed the numbers don’t seem proportional. That’s simply because of the shape of the PQ curve.
In other words, the logarithmic shape of PQ curves uses less code values at the top end of the curve. Only about 7% of the full range is used to define 5,000 – 10,000 nit. Indeed, the PQ curve was designed to ‘spend’ most of the code values in the middle portion of the curve where the argument is they’re more perceptually important.
Video Scopes & HDR – Evolving
One of the reasons I like the Adobe approach so much to HDR Waveform monitoring is that it’s clear and concise and shows you what you need to see.
With that said, I don’t mind how things are in Resolve; I would just like options.
Right now the Resolve team, and in general anyone building scopes for software, probably just aren’t that motivated to build HDR specific graticules – there are just not that many people grading in HDR.
But I think over the next year or two we’ll see that change and see different ways of viewing the HDR signal when it comes to scopes. That change of course is not limited to just the Waveform and RGB parade as larger gamut’s are introduced and more widely used – like REC 2020 color measurement i.e. the vectorscope will probably have to adapt too.
As always questions or additional thoughts please use the comments below.
P.S. After posting this I got several emails about my RED Raw settings and RCM. When switching over to RCM gamma and color space settings become irrelevant BECAUSE of RCM. You can switch these settings to anything you’d like and see no change of the image when in RCM. I only pointed them out to explain what I had done prior to changing the project over to RCM.